Elsevier Editorial System(tm) for International Journal of Multiphase Flow Manuscript Draft Manuscript Number: Title: Microstructural effects on the permeability of periodic fibrous porous media Article Type: Original Research Paper Keywords: Permeability; Fibrous porous media; FEM; Drag relations; Carman-Kozeny equation; Incompressible fluids. Corresponding Author: Mr. Kazem Yazdchi, Corresponding Author's Institution: University of Twente First Author: Kazem Yazdchi Order of Authors: Kazem Yazdchi; S. Srivastava , PhD; S. Luding, PhD Abstract: An analytical-numerical approach is presented for computing the macroscopic permeability of fibrous porous media taking into account their micro-structure. A finite element (FE) based model for viscous, incompressible flow through a regular array of cylinders/fibers is employed for predicting the permeability associated with this type of media. High resolution data, obtained from our simulations, are utilized for validating the commonly used semi-analytical models of drag relations from which the permeability is often derived. The effect of porosity, i.e., volume fraction, on the macroscopic permeability is studied. Also micro-structure parameters like particle shape, orientation and unit cell staggered angle are varied. The results are compared with the Carman-Kozeny (CK) equation and the Kozeny factor (often assumed to be constant) dependence on the micro-structural parameters is reported and used as an attempt to predict a closed form relation for the permeability in a variety of structures, shapes and wide range of porosities.

Professor A. Prosperetti Editor-in-Chief International Journal of Multiphase Flow Department of Mechanical Engineering, The Johns Hopkins University, 119 Latrobe Hall, Baltimore, MD 21218, USA

19th February 2011

Dear professor A. Prosperetti,

Re: Microstructural effects on the permeability of periodic fibrous porous media K. Yazdchi, S. Srivastava and S. Luding Multi Scale Mechanics (MSM), Faculty of Engineering Technology, University of Twente, P.O. Box 217, 7500 AE Enschede, The Netherlands I am submitting the above titled manuscript for review and possible publication in the International Journal of Multiphase Flow journal. Prediction of the permeability of fibrous porous media is a longstanding but still challenging engineering problem, with many practical applications. Here, we consider periodic arrays of parallel cylinders perpendicular to the flow direction and studied with a finite element (FE) based model. We believe that permeability/drag data obtained in this study can be used as a design guide in composites manufacturing and utilized for validation of advanced, more coarse-grained models for particle-fluid interactions.

This manuscript is submitted for publication in International Journal of Multiphase Flow. It has not been previously published, is not currently under review for publication in any other journal and will not be submitted elsewhere during peer review in this journal. I shall be looking forward to hearing from you. Yours sincerely, K. Yazdchi

Cover Letter

1

Microstructural effects on the permeability of periodic fibrous porous media

K. Yazdchi, S. Srivastava and S. Luding

Multi Scale Mechanics (MSM), Faculty of Engineering Technology,

University of Twente, P.O. Box 217, 7500 AE Enschede, The Netherlands

Abstract

An analytical-numerical approach is presented for computing the macroscopic

permeability of fibrous porous media taking into account their micro-structure. A finite

element (FE) based model for viscous, incompressible flow through a regular array of

cylinders/fibers is employed for predicting the permeability associated with this type of

media. High resolution data, obtained from our simulations, are utilized for validating the

commonly used semi-analytical models of drag relations from which the permeability is

often derived. The effect of porosity, i.e., volume fraction, on the macroscopic

permeability is studied. Also micro-structure parameters like particle shape, orientation

and unit cell staggered angle are varied. The results are compared with the Carman-

Kozeny (CK) equation and the Kozeny factor (often assumed to be constant) dependence

on the micro-structural parameters is reported and used as an attempt to predict a closed

form relation for the permeability in a variety of structures, shapes and wide range of

porosities.

Keywords: Permeability; Fibrous porous media; FEM; Drag relations; Carman-Kozeny

equation; Incompressible fluids.

*ManuscriptClick here to view linked References

2

1. Introduction

The problem of creeping flow between solid bodies arranged in a regular array is

fundamental in the prediction of seepage through porous media and has many

applications, including: composite materials [10, 19], rheology [21, 20], geophysics [3],

polymer flow through rocks [27], statistical physics [14, 7], colloid science [26], soil

science [23, 8] and biotechnology [33]. A compelling motivation for such studies

concerns the understanding, and eventually the prediction, of single and multiphase

transport properties of the pore structure.

Usually, when treating the medium as a continuum, satisfactory predictions can be

obtained by Darcy's law, which lumps all complex interactions between the fluid and

fibers/particles into K, the permeability (tensor). Accurate permeability data, therefore, is

a critical requirement for macroscopic simulations based on Darcy’s law – to be

successfully used for design and optimization of industrial processes.

The Ergun equation is a semi-empirical drag relation from which the permeability of

porous media can be deduced. It is obtained by the direct superposition of two asymptotic

solutions, one for very low Reynolds number, the Carman-Kozeny (CK) equation [4],

and the other for very high Reynolds numbers, the Forchheimer correction [4]. However,

these approximations do not take into account the micro-structural effects, namely the

shapes and orientations of the particles, such that not only local field properties but also

some global properties (such as anisotropy) cannot be addressed.

In this respect, two distinct approaches seem to have emerged. The first approach is based

on lubrication theory and considers the pores of a porous medium as a bunch of capillary

tubes which are tortuous or interconnected in a network [4]. Even though this model has

been used successfully for isotropic porous media, it does not work well for either axial

or transverse permeability of aligned fibrous media [5].

The second approach (cell method) considers the solid matrix as a cluster of immobile

solid obstacles, which collectively contribute Stokes resistance to the flow. For a review

of these theories see [10, 4]. When the solids are dilute, i.e., at high porosities, basically

the particles do not feel each other, so that cell approach is appropriate. Bruschke and

Advani [5] used lubrication theory in the high fiber volume fraction range but adopted an

3

analytical cell model for lower fiber volume fractions. A closed form solution, over the

full fiber volume fraction range, is obtained by matching both solutions asymptotically.

Prediction of the permeability of fibrous media dates back to experimental work of

Sullivan [28] and theoretical works of Kuwabara [17], Hasimoto [13], and Happel [12].

The parallel flow solutions are idealized solutions for the flow through cigarette filters,

plant stems and around pipes in heat exchange tanks. The transverse solutions are

applicable to transverse fibrous filters used for cleaning liquids and gases and regulating

their flow. Both types of solutions can also be applicable to the settling of suspensions of

long thin particles. A comprehensive review of experimental works of permeability

calculation of these systems is available in Jackson et al. [15] and Astrom et al. [2].

Later, Sangani and Acrivos [25], performed analytical and numerical studies of the

viscous permeability of square and staggered arrays of cylinders. Their analytical models

were accurate in the limits of low and high porosity. For high densities they obtained the

lubrication type approximations for narrow gaps. Drummond and Tahir [9] modeled the

flow around a fiber using a unit cell approach (by assuming that all fibers in a fibrous

medium experience the same flow field) and obtained equations that are applicable at

lower volume fractions. Gebart [11] presented an expression for the longitudinal flow,

valid at high volume fractions, that has the same form as the well-known KC equation.

For transverse flow, he also used the lubrication approximation, assuming that the narrow

gaps between adjacent cylinders dominate the flow resistance. Using the Eigen-function

expansions and point match methods, Wang [32] studied the creeping flow past a

hexagonal array of parallel cylinders.

Our literature survey indicates that relatively little attention has been paid to the

macroscopic permeability of ordered periodic fibrous materials. More importantly, the

majority of the existing correlations for permeability are based on curve-fitting of

experimental or numerical data and most of the analytical models found in the literature

are not general and fail to predict permeability over the wide range of porosity, since they

contain some serious assumptions that limit their range of applicability.

In this study, periodic arrays of parallel cylinders (with circular, ellipse and square cross-

section) perpendicular to the flow direction are considered and studied with a FE based

model in section 2. The effects of shape and orientation as well as porosity and structure

4

on the macroscopic permeability of the porous media are discussed in detail. In order to

relate our results to available work, the data are compared with previous theoretical and

numerical data for square and hexagonal packing configurations and a closed form

relation is proposed in section 3 in the attempt to combine our various simulations. The

paper is concluded in section 4 with a summary and outlook for future work.

2. Results from FE simulations

This section is dedicated to the FE based model simulations and the results on

permeability as function of porosity, structure, shape and anisotropy.

2.0. Introduction and Terminology

The superficial velocity, U, within the porous media in the unit cell is defined as

uudvV

U

fV

1

, (1)

where u, u , V, Vf and are the local microscopic velocity of the fluid, corresponding

averaged velocity, total volume, volume of the fluid and porosity, respectively. For the

case where the fluid velocity is sufficiently small (creeping flow), the well-known

Darcy’s law relates the superficial fluid velocity U through the pores with the pressure

gradient, p , measured across the system length, L, so that

pK

U

, (2)

where μ and K are the viscosity of the fluid and the permeability of the sample,

respectively. At low Reynolds numbers, which are relevant for most of the composite

manufacturing methods, the permeability depends only on the geometry of the pore

structure.

Recently, models based on Lagrangian tracking of particles combined with computational

fluid dynamics for the continuous phase, i.e., discrete particle methods (DPM), have

become state-of-the-art for simulating gas-solid flows, especially in fluidization processes

5

[16]. In this method, two-way coupling is achieved via the momentum sink/source term,

Sp which models the fluid-particle drag force

p pS u v , (3)

where the interphase momentum-transfer coefficient, , describes the drag of the

gas/fluid phase acting on the particles and vp is the velocity of particles. (Additional

effects like the added mass contributions are disregarded here for the sake of simplicity.)

In steady state, without acceleration, wall friction, or body forces like gravity, the fluid

momentum balance reduces to

0pp u v . (4)

By comparing Eqs. (2) and (4), using the definition of Eq. (1), and assuming immobile

particles, i.e., 0pv , the relation between and permeability K is

K

2 . (5)

Accurate permeability data, therefore, is a critical requirement in simulations based on

DPM to be successfully used in the design and optimization of industrial processes.

In the following, results on the permeability of two-dimensional (2D) regular periodic

arrays of cylinders with different cross section are obtained by incorporating detailed FE

simulations. This is part of a multiscale modeling approach and will be very useful to

generate closure or coupling models required in more coarse-grained, large-scale models.

2.1. Mathematical formulation and boundary conditions

Both hexagonal and square arrays of parallel cylinders perpendicular to the flow direction

are considered, as shown in Fig. 1. The basis of such model systems lies on the

assumption that the porous media can be divided into representative volume elements

(RVE) or unit cells. The permeability is then determined by modeling the flow through

one of these, more or less, idealized cells. At the left and right pressure- and at the top

and bottom periodic-boundary conditions are applied. The no-slip boundary condition is

applied on the surface of the particles/fibers. A typical unstructured, fine and triangular

FE mesh is also shown in Fig. 1(c). The mesh size effect is examined by comparing the

simulation results for different resolutions (data are not shown here). The range of

6

number of elements is varying from 10000 to 100000 depending on the porosity regime.

It should be noted that in Darcy’s linear regime (creeping flow) – although we have

applied pressure boundary conditions at left and right – identical velocity profile at inlet

and outlet are observed, due to the symmetry of this geometry and linearity. However, by

increasing the pressure gradient (data not shown), the flow regime changes to non-linear

and becomes non-symmetric. Furthermore, because of the symmetry in the geometry and

boundary conditions, the periodic boundary condition and symmetry boundary condition,

i.e., zero velocity in vertical direction at top and bottom of the unit cells, will lead to

identical results (as confirmed by simulations – data not shown).

(a) (b)

(c)

Figure 1: The geometry of the unit cells used for (a) square and (b) hexagonal

configurations, with angles 450 and 60

0 between the diagonal of the unit-cell and the

horizontal flow direction (red arrow), respectively. (c) shows a typical quarter of an

unstructured, fine and triangular FE mesh.

Flow direction

7

2.2. Permeability of the square and hexagonal arrays

Under laminar, steady state condition, with given viscosity, the flow through porous

media is approximated by Darcy’s law. By calculating the superficial velocity, U, from

our FE simulations and knowing the pressure gradient, p , over the length of the unit

cell, L, we can calculate the dimensionless permeability (normalized by the cylinder

diameter, d), K/d2. In Table 1, various correlations from the literature are listed. The first

relation by Gebart [11] has a peculiar analytical form and is valid in the limit of high

density, i.e., low porosity – close to the close packing limit c (the same as Bruschke [5]

in the low porosity limit, with maximum discrepancy less than 1%). Note that the

relations by Happel [12], Drummond et al. [9], Kuwabara [17], Hasimoto [13], and

Sangani et al. [25] are all identical in their first term – that is not dependent on the

structure – in the limit of small solid volume fraction , i.e., large porosity. In contrast,

their second term is weakly dependent on the structure (square or hexagonal). Bruschke

et al. [5] proposed relations that are already different in their first term. The last two

relations in the table are only valid in intermediate porosity regimes and do not agree

with any of the above relations in either of the limit cases.

In Fig. 2, the variation of the (normalized) permeability, K/d2, with porosity, for square

and hexagonal packings is shown. The lubrication theory presented by Gebart [11] agrees

well with our numerical results at low porosities ( 0.6 ), whereas, at high porosities

( 0.6 ), the prediction by Drummond [9] better fits our data. Drummond et al. [9] have

found the solution for the Stokes equations of motion for a viscous fluid flowing in

parallel or perpendicular to the array of cylinders by matching a solution outside one

cylinder to a sum of solutions with equal singularities inside every cylinder of an infinite

array. This was in good agreement with other available approximate solutions, like the

results of Kuwabara [17] and Sangani et al. [25] at high porosities, as also confirmed by

our numerical results (data not shown). Note that our proposed merging function in

section 3.4, fits to our FE results within 2% error for the whole range of porosity.

8

0.2 0.4 0.6 0.8 110

-6

10-4

10-2

100

102

K/ d

2

Gebart [11]

Drummond et. al. [9]

FE simulations

0.2 0.4 0.6 0.8 110

-6

10-4

10-2

100

102

K/d

2

Gebart [11]

Drummond et. al. [9]

FE simulations

Figure 2: Normalized permeability plotted against porosity for (a) square and (b)

hexagonal packing for circular shaped particles/cylinders with diameter d, for

perpendicular flow. The lines give the theoretical predictions, see inset. For high

porosities, the difference between Gebart and Drummond, in the hexagonal

configuration, is less than 5%, while for the square configuration it is less than 30%.

(a) Square configuration

(b) Hexagonal configuration

9

Table 1: Summary of correlations between normalized permeability, K/d2 and porosity,

with 1 , the solid volume fraction.

Author K/d2

Comments

Gebart [11]

5/ 2

11

1

cC

4, 1 / 4

9 2

4, 1 / 2 3

9 6

c

c

C

C

Square configuration: 2s

GK d

Hexagonal config.: 2h

GK d

Bruschke

et al. [5]

1

12

2 2

3 2

1tan

1 13 1

12 21

l

l l ll

l l

Lubrication theory, square

config.:

142l

Drummond et

al. [9]

2

2

1 1 2 0.796ln 1.476

32 1 0.489 1.605

2 541 1 2.534

ln 1.497 2 0.73932 2 1 1.2758

Square configuration: 2s

DK d

Hexagonal config.: 2h

DK d

Bruschke

et al. [5]

2

2

1 1 3ln 2

32 2 2

Cell method, square config.

Kuwabara

[17]

21 1ln 1.5 2

32 2

Based on Stokes approximation

Hasimoto

[13]

1 1ln 1.476

32

Using elliptic functions:

21 1ln 1.476 2

32O

Sangani et al.

[25]

2 31 1ln 1.476 2 1.774 4.076

32

Happel [12]

2

2

1 1 1ln

32 1

Lee and Yang

[18]

3

1.3

0.2146

31 1

Valid for 0.435 0.937

Sahraoui et al.

[24]

5.10.0152

1

Valid for 8.04.0

10

2.3. Effect of shape on the permeability of regular arrays

In this subsection, we investigate the anisotropic behavior of permeability due to particle

shape in square configuration. Using elementary algebraic functions, Zhao et al. [34]

derived the analytical solutions for pore-fluid flow around an inhomogeneous elliptical

fault in an elliptical coordinate system. Obdam and Veling [22] employed the complex

variable function approach to derive the analytical solutions for the pore-fluid flow within

an elliptical inhomogeneity in a two-dimensional full plane. Zimmerman [36] extended

their solutions to a more complicated situation, where a randomly oriented distribution of

such inhomogeneous ellipses was taken into account. Wallstrom et al. [31] later applied

the two-dimensional potential solution from an electrostatic problem to solve a steady-

state pore-fluid flow problem around an inhomogeneous ellipse using a special elliptical

coordinate system. More recently, Zhao et al. [35] used inverse mapping to transform

those solutions into a conventional Cartesian coordinate system.

Here, in order to be able to compare different shapes and orientations, the permeability

will be normalized with respect to the obstacle length, Lp, which is defined as

Lp = 4 area / circumference

Lp = 2r = d (for circle), Lp = c (for square), Lp = 4πab / AL (for ellipse) (6)

where r, c, a and b are the radius of the circle, the side-length of the square, the major

(horizontal) and minor (vertical) length of the ellipse, respectively. AL is the

circumference of the ellipse.

By applying the same procedure as in the previous section, the normalized permeability

(with respect to obstacle length, Lp) is calculated for different shapes on a square

configuration.

In Fig. 3 the normalized permeability is shown as function of porosity for different

shapes. At high porosities the shape of particles does not affect much the normalized

permeability, but at low porosities the effect is more pronounced. We observe that circles

have the lowest and horizontal ellipses the highest normalized permeability.

11

0.2 0.4 0.6 0.8 110

-6

10-4

10-2

100

102

K/L

p 2

Circle

Square

Ellipse

Figure 3: Effect of shape on the normalized permeability from a square packing

configuration of circles, squares and ellipses (a/b=2, major axis in flow direction). The

lines are only connecting the data-points as a guide to the eye.

2.4. Effect of aspect ratio on the permeability of regular arrays of ellipses

In this subsection the effect of aspect ratio, a/b on the normalized permeability of square-

arrays of ellipses is investigated. In fact, the case of high aspect ratio at high porosity

represents the flow between parallel plates (slab flow). The relation between average

velocity, u , and pressure drop for slab flow is

2

12

sh pu

L

(7)

where hs is the distance between parallel plates (in our square configuration, in the limit

/ 1a b , one has hs=L=1). Note that, since there are no particles, 1 , the average and

superficial velocities are identical, i.e., u U . By comparing Eqs. (7) and (2) the

permeability, i.e., 2 /12sK h is obtained, which indeed shows the resistance due to no

slip boundaries at the walls. The variation of permeability for a wide range of aspect

ratios at different porosities is shown in Fig. 4. It is observed (especially at high

12

porosities) that by increasing the aspect ratio the permeability increases until it reaches

the limit case of slab flow for which the permeability is 2 /12 0.0833sK h [m

2]. The

aspect ratio b/a <1 means that the ellipse stands vertically and therefore the permeability

reduces drastically.

10-2

10-1

100

101

102

10-8

10-6

10-4

10-2

100

b/a

K [m

2]

= 0.6

= 0.7

= 0.8

= 0.9

= 0.95

Slab flow (hs

2 /12 [m

2])

Figure 4: Effect of aspect ratio on the permeability of square configurations of ellipses

with different porosities as given in the inset. The lines are only connecting the data-

points as a guide to the eye.

2.5. Effect of orientation on the permeability of regular arrays

By changing the orientation (θ), i.e., the angle between the major axis of the obstacle and

the horizontal axis, not only the values of the permeability tensor will change, but also its

anisotropy will show up (so that the pressure gradient and the flow velocity are not

parallel anymore). Therefore, the geometry of the pore structure has great effect on the

permeability in irregular fibrous media. This effect for squares and ellipses (a/b = 2) is

shown in Fig. 5.

13

For square shapes, at high porosities, the orientation does not much affect the

permeability, whereas at low porosities the permeability depends a lot on the orientation.

At θ = 450 we observe a drop in permeability, because we are close to the blocking

situation, i.e., zero permeability, at a critical porosity (at which the permeability drops to

zero) of 2

11 0.5

2sin 45c

.

For ellipses, at high porosity, the orientation does not affect the permeability, whereas at

low porosities the effect is strong. By increasing the orientation angle, i.e., by turning the

major axis from horizontal to vertical, the permeability is reduced. The critical porosity

εc≈0.6073 is purely determined by the major axis of the ellipse for θ = 900.

14

0.2 0.4 0.6 0.8 110

-5

10-4

10-3

10-2

10-1

100

101

K/L

p 2

= 0o

= 10o

= 20o

= 30o

= 45o

(a)

0.2 0.4 0.6 0.8 1

10-4

10-5

10-3

10-2

10-1

100

101

K/L

p 2 = 0o

= 30o

= 45o

= 60o

= 90o

(b)

Figure 5: Effect of orientation (θ) on the normalized permeability for (a) square and (b)

ellipse (a/b=2) in square packing configurations at different porosity. The lines are only

connecting the data-points as a guide to the eye.

15

The general form of Darcy’s law for anisotropic media in 2D in matrix form can be

written as

11 12

21 22

1x

y

pU K K x

pK KU

y

, (8)

where <Ux> and <Uy> are superficial velocities in x and y direction, respectively. Then

the permeability tensor for any value of can be calculated as

011 12

9021 22

0

0

TKK K

K R RKK K

(9)

where 0K and 90K are the principal values of permeability that are determined from the

values of 00 and 090 , respectively. In Eq. (9), RT is the transposed of the rotation

matrix R, defined as (counterclockwise rotation by )

cossin

sincosR (10)

Eq. (9) shows that for 00 90,0 , one has 012 K , which means that by applying a

pressure gradient in x direction, one gets a superficial velocity in y direction (i.e.

anisotropic behavior because of oriented shape). Our numerical results are in good

agreement with theoretical predictions (Eq. (9)) especially at high porosities (see the solid

lines in Fig. 6). We have more deviation at low porosities (maximum discrepancy ≈ 5%)

because of channel blockage and changes in flow behavior (the comparison is not shown

here).

In Fig. 6, the variation of normalized permeability is shown as a function of the

orientation angle . The normalized permeability is symmetric to 900 and decreases by

increasing the orientation angle from 00. The eigenvalues of the permeability tensor are

the extrema of the curves and the other data are well fitted by

2

0 90 0 902 cos 2 2pK L K K K K . By decreasing the aspect ratio a/b, we

approach the value for the circular (cylinder) obstacle, i.e., a/b =1. The normalized

permeability is symmetric to 450 for square shapes.

16

0 20 40 60 80 100 120 140 160 1800

0.05

0.1

0.15

0.2

0.25

0.3

0.35

[degree]

K/L

p2

Figure 6: Normalized permeability plotted against orientation angle for different shapes

at porosity 85.0 , for different obstacle shapes in square configurations. The dashed

line represents circles. The solid lines show the theoretical predictions according to Eq.

(9), where the eigenvalues are taken from the 00, 90

0 and 0

0, 45

0 degrees for ellipses and

squares simulations, respectively.

2.6. Effects of staggered cell angle

In this subsection, the effect of another micro-structural parameter, the staggered cell

angle, , on the normalized permeability for circles (Lp=d) is discussed. The staggered

angle is defined between the diagonal of the unit-cell and flow-direction (horizontal), see

Fig. 7. In addition to the special cases o45 and o60 , i.e., square and hexagonal

packings, respectively, several other angles are studied. The contour of the horizontal

velocity field component, for different , at constant porosity 0.7 , is shown in Fig.

7. By changing , the flow path and also the channel length will change. At 070 and

higher, the flow mainly follows a straight line, indicated by arrows in Fig. 7(a), with large

superficial velocity and consequently large values of permeability. However, by

decreasing down to 350, the flow pattern completely changes and the superficial

17

velocity reduces, which should lead to lower and lower permeability. In brief, with

increasing angle, both the superficial velocity and the permeability increase, with a

plateau around o45 .

(a) (b)

(b) (d)

Figure 7: Horizontal velocity field components for (a) 060 , (b)

050 , (c)

040 and (d) 035 at fixed porosity ε = 0.7. The arrows indicate the main flow

channel in (a) and (b). The staggered angle is defined between the diagonal of the unit-

0 0.03 0.06 0.09 0.12 0.15 0.18 0.21 0.24 [m/s]

18

cell and flow-direction (horizontal). The minimal angle 1 0

min

2 1tan 10.81

is realized when the vertical opening is closed, while the maximal angle

1 0

max tan 79.182 1

corresponds to the closed horizontal pore.

In Fig. 8 the normalized permeability is shown as function of the staggered angle, , at

different porosities. As it is seen the arrangement of particles relative to the flow

direction is important in determining the permeability. By increasing , the normalized

permeability increases (the vertical distance between particles increases and therefore the

resistance to the flow decreases) until it reaches a local maximum at 035 –

consistently for different porosities. At larger angles, it slightly decreases and attains a

local minimum at 055 , beyond which it increases rapidly again. This behavior can be

explained by the variation of the area-fraction distribution with on the planes

perpendicular to the flow direction, as discussed in [1].

The normalized permeability as a function of can be expressed as a cubic polynomial

3 20 0 0

2 0 0 0

45 45 45A B C D

45 45 45

K

d

(11)

where A, B, C and D are constants, listed in Table 2 for different porosities.

19

20 30 40 50 60 700

0.02

0.04

0.06

0.08

0.1

0.12

0.14

[degree]

K/d

2

Figure 8: The variation of normalized permeability versus staggered unit cell angle at

different porosities, as given in the inset. The solid lines show the fit (Eq. (11))

Table 2: Fitted parameters for the permeability-staggered angle relation

8.0 7.0 6.0

A 0.38401 0.18813 0.09913

B 0.02682 0.01738 0.01171

C -0.03693 -0.01914 -0.01027

D 0.07698 0.02584 0.009319

The leading term with A is dominating, the term with B is a rather small correction, the

term with C sets the (negative) slope in the center, and the term with D determines the

offset. All fit-parameters depend on porosity and it should be noted that the range of

available angles is limited and also depends on porosity. Additional scaling- and fit-

attempts (data not shown) did not lead to much better results, thus we only present this

empirical fit here.

The decreasing region, i.e., 0 035 55 corresponds to the case in which the flow goes

in a preferred channel orthogonal/perpendicular to the line (diagonal) connecting two

particles, see Figs. 7(b) and (c). While in cases with larger , the flow goes at straight

20

lines/channels, see Fig. 7(a), the configuration for smaller is dominated by the narrow

vertical opening between two obstacles. In essence, in the plateau region, the

permeability is not much affected by the staggered angle . This observation might be

useful during design and manufacturing of fibrous composites.

In summary, the results of this section show that the macroscopic permeability not only

depends on the porosity but also on the microstructure, namely shape, aspect ratio and

orientation of particles.

3. Theoretical prediction of the permeability for all porosities

In this section, based on the observations in the previous section, and using the velocity

profiles from FE unit cell simulations, a generalized form of the Carman Kozeny (CK)

equation for the permeability of fibrous porous media (2D regular arrays of particles) is

proposed.

3.1 From special cases to a more general CK equation

The earliest and most widely applied approach in the porous media literature, for

predicting the permeability, involves capillary models [6] such as the one that leads to the

CK equation. The approach to obtain this equation is based on Poiseuille flow through

pipes. Assuming pipe flow through a cylindrical channel of diameter hp, the average

velocity through the channel is

2

32

p

p

h pu

L

, (12)

and for slab flow through an infinite channel of width hs

2

12

ss

h pu

L

, (13)

given the pressure drop p per length L, and a fluid with viscosity .

Defining the hydraulic diameter

volume available for flow4

total wetted surfacehD , (14)

21

allows to unify the relations above, by combining Eq. (14) with either Eq. (12), with

h pD h , or Eq. (13), with 2h sD h , and with Darcy’s law, Eq. (2), so that the

permeability is described by the CK relation [6]

2

h

CK

u DK

p

(15)

where CK =32 (or 48) is the dimensionless Kozeny factor, characteristic of the pipe (or

slab) pore structure. When one has obstacles like fibers (or particles) instead of straight

pores, the hydraulic diameter can be re-written as

4 4 particle surface 4

, with1 1 particle volume 1

vh v

v v

SV dD a

S a V d

, (16)

with the total volume of the unit cell, V, the total wetted surface, Sv, the specific surface

area, av, and the porosity, , for a fibrous medium of fiber diameter d. Note that the

hydraulic diameter, in this way, is expressed as a function of the measurable quantities

porosity and specific surface area. The above value of av is for circles (cylinders) – for

spheres one has av=6/d. In this formulation, we just consider the resistance due to

presence of particles (no slip boundaries at the surface of the particles) and neglect the

outer walls.

Inserting Eq. (16) into Eq. (15), yields the normalized permeability for fibers

3

22

1

1CK

K

d

, (17)

which depends non-linearly on the porosity and on a shape/structure factor, CK .

One of the main drawbacks of the CK equation is that the Kozeny factor CK is a-priori

unknown in realistic systems and has to be determined experimentally. An ample amount

of literature exists on the experimental and theoretical determination of the Kozeny

factor, but we are not aware of a theory that relates CK with the microstructure, i.e., the

porosity, the random configuration, tortuosity, and other microscopic quantities. An

overview of experimental and theoretical approaches can be found in Ref. [2], which

mainly deals with fibrous media, and Ref. [29], which is based on variational principles.

One of the most widely accepted approaches to generalize the CK relation was proposed

22

by Carman [6], who noticed that the streamlines in a porous medium are far from being

completely straight and parallel to each other. This effect can be described by a

dimensionless parameter, Le/L (tortuosity), with the length of the streamlines, Le, relative

to the length of the sample, L. Hence the Kozeny factor can be split into

2

eCK

L

L

(18)

where is the effect of particle shape, which can be seen as a fitting parameter. In fact,

the tortuosity and the shape factor reflect the effects of microstructure on the macroscopic

properties (like permeability) of the porous media.

In the original form of the CK equation for random 3D sphere structures, it is assumed

that the tortuosity is a constant for all ranges of porosities and is equal to 2 and the

fitting parameter, , then becomes 90 for the case of pipe flow and 60 for slab flow.

Knowing the values of the normalized permeability from our FE simulations, we can

compare the values of the Kozeny factor based on our numerical results and available

theoretical data and with the original CK =120 for slab flow, see Eqs. (17) and (18). The

comparison is shown in Fig. 9. At a certain range of porosities, 7.05.0 , the CK

factor CK is indeed not varying much. However, at higher or lower porosities, it

strongly depends on porosity and structure. At high and low porosities, our numerical

results are in good agreement with the predictions of Drummond [9] and Gebart [11],

respectively (see Table 1). These results indicate that the Carman-Kozeny factor, CK , is

indeed not constant and depends on the microstructure.

In the following subsections, we will study the dependency of CK on the micro-

structural parameters.

23

0.2 0.4 0.6 0.8 10

100

200

300

400

500

600

K

C

Original CK

FE simulation (square)

FE simulation (hexagonal)

Gebart [11] (hexagonal)

Gebart [11] (square)

Drummond [9] (hexagonal)

Drummond [9] (square)

Figure 9: Kozeny factor plotted as a function of porosity for different models (lines) and

data sets (symbols) as given in the inset.

3.2. Measurement of the tortuosity (Le/L)

As discussed before, the tortuosity is the average effective streamline length scaled by

system length, Le/L, and one possible key parameters in the Kozeny factor in the CK

equation [6]. From our numerical simulations, we extract the average length of several

streamlines (using 8 streamlines that divide the total mass in-flux into 9 zones, thus

avoiding the center and the edges). By taking the average length of these lines, the

tortuosity can be obtained, while by taking the standard deviation of the set of

streamlines, the homogeneity of the flow can be judged. The tortuosity is plotted in Fig.

10 as function of porosity for different shapes and orientations and as function of the

staggered angle for different porosities.

Unlike the traditional form of the CK equation, which assumes that / 2eL L (for

random 3D structures) is constant [6], our numerical results show that the tortuosity (i) is

smaller and (ii) depends on the porosity and the pore structure. In Fig.10(a), as intuition

suggests, the vertical and horizontal ellipses have the highest and lowest average

24

tortuosity, respectively. This goes ahead with very large and very small standard

deviation, i.e., the vertical ellipse configuration involves the widest spread of streamline

lengths. In the case of a circular shape obstacle, the (average) tortuosity is between the

horizontal and vertical ellipse cases and, for intermediate porosity, becomes almost

independent of porosity, with constant standard deviation that is wider than the average

tortuosity deviation from unity. The square shape obstacles are intermediate in tortuosity,

i.e., the 00-square (45

0-square) shapes take tortuosity values between the horizontal

(vertical) ellipse and the circular obstacles.

In Fig. 10(b), the tortuosity is plotted against the staggered unit cell angle . By

increasing the value of tortuosity increases until about 045 , where it reaches its

maximum. Note that the standard deviation remains small for all angles smaller than

045 . At higher angles, tortuosity decreases, while its standard deviation considerably

increases. At large values of , most of the fluid flow goes along a straight line,

however, near the boundary, we have a few longer streamlines that cause the large

standard deviation. At the limit case of max , when the upper particles touch, see Fig.

7, the tortuosity approaches unity (data not shown) and the flow goes mostly along a

straight channels.

3.3. Measurement of the shape/fitting factor ( )

Knowing the values of tortuosity, Le/L, and normalized permeability, K/d2, from our FE

simulations, we can obtain the values of for different shapes and orientations as

2

eCK

L

L

. In Fig. 11 the variation of as a function of porosity for different

shapes and orientations is shown. Unlike the traditional CK factor, the shape/fitting factor

is not only a function of porosity but also depends on the orientation of

particles/cylinders. This dependency is more pronounced at high (low) porosities and

close to the blocking conditions, i.e., ellipses with 900 and squares with 45

0.

25

0.4 0.5 0.6 0.7 0.8 0.9 10.98

1

1.02

1.04

1.06

1.08

1.1

1.12

1.14

1.16

Le/L

Ellipse-90o

Square-45o

Circle

Square-00

Ellipse-00

(a)

10 20 30 40 50 60 70 801

1.05

1.1

1.15

1.2

1.25

[degree]

Le/L

= 0.6

= 0.7

= 0.8

(b)

Figure 10: Tortuosity LLe / (a) plotted as a function of porosity for different obstacle

shapes and orientations on square configurations, o45 , (b) plotted as a function of the

staggered cell angle, , as given in the inset, for circles at different porosities on

hexagonal configurations. Error bars give the standard deviation of the 8 streamline

26

lengths, where bottom-values below unity indicate a highly non-symmetric distribution

around the average.

0.4 0.6 0.8 10

500

1000

1500

2000

Ellipse-90

0

Square-450

Circle

Square-00

Ellipse-00

Figure 11: Shape/fitting factor, , plotted as a function of porosity for different obstacle

shapes and orientations on square configurations, o45 . The straight dashed line

shows the value of 60 as in the original CK factor.

3.4. Corrections to the limit theories

Being unable to explain the variation of permeability with tortuosity and a constant shape

factor, now we attempt to optimize/correct the limit theories by Gebart [11] and

Drummond [9], see Table 1, in order to propose an analytical relation for the permeability

that is valid for all porosities and for square and hexagonal arrays of cylinders.

3.4.1 Square configuration

Assuming one particle at the center, pressure boundary at the left and right and periodic

top and bottom, we correct the original Gebart relation from Table 1, s

GK , by a linear

correction term 2

2

1

1

s s

G G

c

K Kg

, with g2=0.336. In contrast to s

GK , which

asymptotically approaches the limit case, c , but for 0.6 deviates already by

27

about 10%, the correction, KG2, has a maximum discrepancy in the range 0.85c of

less than 10%, and for 0.7c of less than 2%, see the squares in Fig. 12.

Since the Drummond relation from Table 1, s

DK , is valid at high porosities with

maximum discrepancy at 0.7 1 of less than 10% and for 0.8 1 of less than 2%,

we propose the following merged globally valid (for all porosities) function

2 2 ,s s s s

G D GK K K K m with 1 tanh /

,2

h tm

0.75, 0.037h t ,

that is valid for the whole range of porosity with deviations of less than 2% and it

includes the analytical relations for the limit cases.

0.2 0.4 0.6 0.8 10.6

0.7

0.8

0.9

1

1.1

FE

M / fit

KG

s [11]

Ks

G2

KD

s [9]

Tanh

Figure12: Relative error between FEM results and proposed corrections for square

configuration with the critical porosity 1 / 4c .

3.4.2 Staggered hexagonal configuration 060 :

In this situation, the correction to the Drummond relation from Table 1 is

2 1 21h h

D DK d K d , 2 ,h h h h

G D GK K K K m with d1=0.942, d2=0.153,

0.55, 0.037h t , leads to a corrected permeability for all porosities, with a

maximum error of less than 2%, see Fig. 13.

28

0.2 0.4 0.6 0.8 10.6

0.7

0.8

0.9

1

1.1

FE

M / fit

KG

h [11]

KD

h [9]

KD2

h

Tanh

Figure 13: Relative error between FEM results and proposed fits for hexagonal

configuration with the critical porosity 1 / 2 3c .

4. Summary and Conclusions

The permeability of porous structures is an important property that characterizes the

transport properties of porous media; however, its determination is challenging due to its

complex dependence on the microstructure of the media. Using an appropriate

representative volume element, transverse flow in aligned, periodic fibrous porous media

has been investigated based on high resolution (fine grid) FE simulations. This is

complementary to the recent study by Van der Hoef et. al. [30] who obtained the

drag/permeability relation for random arrays of monodisperse and bidisperse spheres. In

all of our simulations, the total pressure drop has been chosen small, such that we are

always in Darcy’s regime (creeping flow). In particular, the effects of different

parameters including fiber (particle) shape, aspect ratio, orientation, and staggered unit

29

cell angle on the normalized permeability are measured and discussed in detail for the full

range of porosities. The conclusions are:

The present results for the permeability are validated by comparing with available

theoretical and numerical data for square and hexagonal arrays over a wide range

of porosities. Especially in the limits of high and low porosity, agreement with

previous theoretical results is established.

By increasing the staggered unit cell angle, (where 60 degrees corresponds to

the hexagonal array), from the blocked configuration with minimal angle min, the

normalized permeability increases until it reaches its local maximum at 035 .

Then it decreases a bit (almost plateau) until it reaches its local minimum at 055 . From there it increases again until a maximum porosity is reached at

max. The best-fit (3rd

order) polynomials at different porosities are presented as

reference for later use.

By increasing the orientation angle of the ellipses (here the longer axis of ellipses

was used to define the orientation relative to the flow direction), the permeability

decreases and shows its anisotropy. The permeabilities at the extreme cases, i.e.,

eigenvalues at 00 and 90

0, are used to predict the permeability for arbitrary

orientation angles, see Eq. (9).

By increasing the aspect ratio of horizontal ellipses, the permeability increases

and approaches the permeability of slab flow, i.e., 2 /12 0.0833sK h [m

2] at

high porosities.

Using the hydraulic diameter concept the permeability can be expressed in the

general form of the (Carman Kozeny) CK equation. Our numerical results show

that the CK factor not only depends on the porosity but also on the pore structure,

namely particle shape, orientation and staggered angle .

The numerical results show that immobile circles and ellipses have the lowest and

highest permeability, respectively. The relevance of this observation for flows of

gas/fluid-solid with moving non-spherical particles is an open question.

Since analytical forms with the power as a free fit-parameter are neither consistent with

the highest porosity asymptote, nor with the lowest porosity limit case, those fits are only

an attempt to describe the intermediate regime of practical importance with a closed

functional form. In order to improve the analytical relation for the permeability, to be

applied, e.g., for DEM-FEM coupling, we proposed a merged function that includes both

30

limit cases of low and high porosity and is smooth in between with maximal deviation

from our numerical results of less than 2%.

Future work will investigate the creep and inertial flow regime through periodic and

disordered arrays, the relation between microstructure and (macro) permeability, and the

effect of the size of the system, especially for random/disordered structures. Since already

the packing generation algorithm affects the permeability in random arrays of parallel

cylinders, different procedures have to be compared and evaluated with respect to the

microstructure. These results can then be utilized for validation of advanced, more

coarse-grained models for particle-fluid interaction and their coupling-terms between the

discrete element method (DEM) for the particles and the FEM or CFD solver for the

fluid, in a multi-scale coarse grained approach.

Acknowledgements:

The authors thank N. P. Kruyt, A. Thornton and T. Weinhart for helpful discussion and

acknowledge the financial support of STW through the STW-MuST program, project

number 10120.

References

[1] Alcocer, F.J., Kumar, V. and Singh, P., 1999. Permeability of periodic porous

media, Physical Review E 59(1), 711-14.

[2] Astroem, B., Pipes, R., Advani, S., 1992. On flow through aligned fiber beds

and its application to composite processing, Journal of Composite Materials 26

(9), 1351– 73.

[3] Berryman, J.G., Wang, H.F., 2000. Elastic wave propagation and attenuation in

a double-porosity dual-permeability medium, Int. J. Rock Mech. and Mining

Sci. 37, 67-78.

[4] Bird, R.B., Stewart, W.E. and Lightfoot, E.N., 2001. Transport Phenomena, 2nd

edn,. John Wiley & Sons.

31

[5] Bruschke, M.V. and Advani, S.G., 1993. Flow of generalized Newtonian fluids

across a periodic array of cylinders, Journal of Rheology 37, 479-98.

[6] Carman, P.C., 1937. Fluid flow through granular beds, Transactions of the

Institute of Chemical Engineering 15, 150–66.

[7] Chen, X., Papathanasiou, T.D., 2008. The transverse permeability of disordered

fiber arrays: A statistical correlation in terms of the mean nearest interfiber

spacing, Transport in Porous Media 71 (2), 233-251.

[8] Crawford, J.W., Matsui, N., Young, I.M., 1995. The relation between the

moisture-release curve and the structure of soil, Euro J. Soil Sci. 46, 369-375.

[9] Drummond, J.E. and Tahir, M.I., 1984. Laminar viscous flow through regular

arrays of parallel solid cylinders, Int. J. Multiphase Flow 10, 515-40.

[10] Dullien, F.A.L., 1992. Porous Media: Fluid Transport and Pore Structure, 2nd

edn. Academic Press, New York.

[11] Gebart, B.R., 1992. Permeability of Unidirectional Reinforcements for RTM,

Journal of Composite Materials 26, 1100–33.

[12] Happel, J., 1959. Viscous flow relative to arrays of cylinders, AIChE 5, 174– 7.

[13] Hasimoto, H., 1959. On the periodic fundamental solutions of the Stokes

equations and their application to viscous flow past a cubic array of spheres, J.

Fluid Mech. 5, 317–328.

[14] Hilfer, R., 2000. Local porosity theory and stochastic reconstruction for porous

media, in: K. Mecke and D. Stoyan (eds.), Statistical Physics and Spatial

Statistics, Lecture Notes in Physics, Springer, Berlin.

[15] Jackson, G.W., James, D.F., 1986. The permeability of fibrous porous media,

Can. J. Chem. Eng. 64, 364–374.

[16] Kuipers, J.A.M., Van Duin, K.J., Van Beckum, F.P.H., Van Swaaij, W.P.M.,

1993. Computer simulation of the hydrodynamics of a two-dimensional gas-

fluidized bed, Computers Chem. Eng. 17, 839-858.

[17] Kuwabara, S., 1959. The forces experienced by randomly distributed parallel

circular cylinders or spheres in a viscous flow at small Reynolds numbers,

Journal of the Physical Society of Japan 14, 527–532.

32

[18] Lee, S.L., Yang, J.H., 1997. Modeling of Darcy–Forchheimer drag for fluid

flow across a bank of circular cylinders, Int. J. Heat Mass Transfer 40, 3149–55.

[19] Laakkonen, K., 2003. Method to model dryer fabrics in paper machine scale

using small scale simulations and porous media model, Int. J. Heat Fluid Flow

24, 114–121.

[20] Moss, G.R., Rothstein, J.P., 2010. Flow of wormlike micelle solutions through a

periodic array of cylinders, Journal of Non-Newtonian Fluid Mechanics 165 (1-

2), 1-13.

[21] Muller, A.J. and Saez, A.E., 1999. The Rheology of Polymer Solutions in

Porous Media, Flexible Polymer Chains in Elongational Flow, Theories and

Experiments, Springer-Verlag, Heidelberg.

[22] Obdam, A.N.M., Veling, E.J.M., 1987. Elliptical inhomogeneities in

groundwater flow-An analytical description, Journal of Hydrology 95, 87–96.

[23] Roth, K., 2008. Scaling of water flow through porous media and soils, Euro. J.

Soil Sci. 59 (1), 125-130.

[24] Sahraoui, M., Kaviany, M., 1992. Slip and no-slip boundary conditions at

interface of porous, plain media, Int. J. Heat Mass Transfer 35, 927–43.

[25] Sangani, A.S., Acrivos, A., 1982. Slow flow past periodic arrays of cylinders

with application to heat transfer, Int. J. Multiphase Flow 8, 193–206.

[26] Shani, C., Weisbrod, N., Yakirevich, A., 2008. Colloid transport through

saturated sand columns: Influence of physical and chemical surface properties

on deposition, Colloids and Surfaces A: Physicochemical and Engineering

Aspects 316 (1-3), 142-150.

[27] Sorbie, K.S., Parker, A., Clifford, P.J., 1987. Experimental and theoretical study

of polymer flow in porous media, SPE Reservoir Eng. 2, 281-304.

[28] Sullivan, R.R., 1942. Specific surface measurements on compact bundles of

parallel fibers, J. Appl. Phys. 13, 725–730.

[29] Torquato S., 1991. Random heterogeneous media: Microstructure and improved

bounds on effective properties, Appl. Mech. Rev. 44, 37-77.

[30] Van der Hoef, M.A., Beetstra, R., Kuipers, J.A.M., 2005. Lattice-Boltzmann

simulations of low-Reynolds-number flow past mono- and bidisperse arrays of

33

spheres: Results for the permeability and drag force, Journal of Fluid Mechanics

528, 233-254.

[31] Wallstrom, T.C., Christie, M.A., Durlofsky L.J., Sharp D.H., 2002. Effective

flux boundary conditions for upscaling porous media equations, Transport in

Porous Media 46, 139–153.

[32] Wang, C.Y., 2002. Slow Viscous Flow between Hexagonal Cylinders,

Transport in Porous Media 47, 67–80.

[33] Wanner, O., Cunningham, A.B., Lundman, R., 1995. Modeling biofilm

accumulation and mass transport in a porous medium under high substrate

loading, Biotechnology and bioengineering 47, 703-712.

[34] Zhao, C., Hobbs, B.E., Ord, A., Peng, S., Liu, L., Mühlhaus H.B., 2006a.

Analytical solutions for pore-fluid flow focusing within inclined elliptic

inclusions in pore-fluid-saturated porous rocks: Solutions derived in an elliptical

coordinate system. Mathematical Geology 38, 987–1010.

[35] Zhao, C., Hobbs, B.E., Ord, A., Peng, S., Liu, L., 2008. Inversely-Mapped

Analytical Solutions for Flow Patterns around and within Inclined Elliptic

Inclusions in Fluid-Saturated Rocks, Mathematical Geosciences 40, 179–197.

[36] Zimmerman, R.W., 1996. Effective conductivity of a two-dimensional medium

containing elliptical inhomogeneities, Proceedings of the Royal Society A:

Mathematical, Physical and Engineering Sciences 452, 1713–27.